1. Valentina Matovic, Mašinski fakultet Univerziteta u Beogradu, Serbia
2. Jasna Trbojević-Stanković, Medicinski fakultet Beograd, Serbia
3. Lidija Matija, Masinski fakultet, Univerzitet u Beogradu, Serbia
4. Dušan Šarac, Serbia
5. Aleksandra Vasić-Milovanović, Masinski fakultet, Univerzitet u Beogradu, Serbia
6. Andrija Petrović, Masinski fakultet, Univerzitet u Beogradu, Serbia
7. Nikola Stojiljković, Mašinski fakultet Univerziteta u Beogradu, Serbia
Despite the prevalent use of recombinant human erythropoietin, anemia is a frequent finding in hemodialysis (HD) patients. The major cause of anemia in chronic renal disease is the lack of erythropoietin, which is the main stimulant of erythropoiesis. The prerequisite for the successful treatment is sufficient amount of iron that is incorporated into erythrocytes. Near-infrared spectroscopy (NIRS) is applied as a non-invasive on-line detection method of blood iron level from the spent HD effluent. The blood iron levels were presented in the form of a binomial variable, where 0 indicates an iron level below the normal limit (below 8 mmol/l) and 1 represents iron level within the reference range (8-30 mmol/l). We used Machine Learning (ML) algorithms: Random Forest (RF), Logistic Regression (LR), K-nearest neighbor (KNN), Support Vector Machine (SVM), Decision Tree Classifier (DT), and Gaussian Naive Bayes (NB) to classify blood iron levels. Area Under the Curve (AUC) and accuracy were utilized for model evaluation.
RF and KNN have shown the best classification accuracy in predicting blood iron level, while SVM, DT, NB and LR, showed average accuracy. AUC score of RF algorithm was 90% with an accuracy of 97%, and AUC score of KNN algorithm was 86% with an accuracy of 85%.
The NIRS with ML methods accurately predict iron levels in HD patients and thus can be used as a non-invasive assessment method.
SIMPOZIJUM B - Biomaterijali i nanomedicina